Digit string experiment -- instructions for the experimenter

1. User instructions and strings to be read:

English:
http://stronsay.ldc.upenn.edu/myl/Es1b.html
http://stronsay.ldc.upenn.edu/myl/Es2b.html
http://stronsay.ldc.upenn.edu/myl/Es3b.html
http://stronsay.ldc.upenn.edu/myl/Es4b.html
http://stronsay.ldc.upenn.edu/myl/Es5b.html

Hungarian:
http://stronsay.ldc.upenn.edu/myl/Hs1b.html
http://stronsay.ldc.upenn.edu/myl/Hs2b.html
http://stronsay.ldc.upenn.edu/myl/Hs3b.html
http://stronsay.ldc.upenn.edu/myl/Hs4b.html
http://stronsay.ldc.upenn.edu/myl/Hs5b.html

(Other languages may be added later.)

These all have 100 4-digit strings, where each digit occurs equally often in each position in the string, and each pair of digits occurs equally often spanning each pair of string positions. The different files are just different random lists meeting those conditions.

For the pilot experiment, I suggest that we all use the Es2b.html or Hs2b.html lists.

You may find that it works best to print out the instructions and the list of strings to be read; if so, you may want to format them so that the instructions are on a separate page.

2. Speaker information:

From each speaker, get the following demographic information:

Age (in years); sex (M or F); educational level (highest degree); place of birth; other places of long-term residence.

For me, the information would be: 59; M; PhD; Connecticut (USA); Massachusetts, New Jersey, Pennsylvania.

This information will of course be kept confidential.

3. How to make the recording.

A perfect recording is not necessary, but there is also no reason to make it unnecessarily hard to measure.

Record in a reasonably quiet place, using a microphone placed near to the speaker's mouth (while avoiding the having the speaker's breath directly hit the microphone.). A head-mounted microphone is fine if you have one. Recording directly to your computer is usually the easiest thing.

Check the signal level, so it is not too low but also isn't clipped.

A monophonic (single-channel) recording is best, and a sampling rate of 22050 or 11025 or even 8000 Hz is fine. (44100 or 48000 is fine too, but the longer files will take longer to upload, and the extra frequency range is not necessary in this case.)

If you aren't already familiar with making digital recordings, you might want to download and use a few software application such as WaveSurfer or Audacity:

http://www.speech.kth.se/wavesurfer/
http://audacity.sourceforge.net/

These will allow you to set the type of recording (number of channels, sampling rate), and check the recording level. Let me know if you need help with this. (Unfortunately, there is still a lot of computer-to-computer variation in how devices are chosen and levels are set, etc., so I won't try to give one-size-fits-all instructions here.)

If you're familiar with making digital recordings, feel free to use the methods that you're used to. If you have any doubts about whether your recording methods are going to be appropriate for this purpose, please ask.

4. Monitoring the recording

You may find that it's helpful to have the speaker read the first few items for practice, before starting the recording in earnest.

While the speaker is reading the list, follow along and note any problems. If they make mistakes or produce disfluent examples, ask them to repeat the relevant list items.

5. Uploading the recordings

We'll want to have the recording measured in the same way, to ensure consistency, so I'll ask you to send me a copy. (This also means less work for you -- though if you want to participate in segmentation and analysis of the data, you'll be welcome to do so.)

When you're done with the recordings, let me know and I'll send instructions for uploading them. We'll start with the idea of using http://www.sublimation.org/scponly/ to a server here at Penn, which will mean using scp or WinSCP or the like on your end -- if we run into problems, we'll try other methods.